GE Effects of Cash Transfers: Pre-analysis plan for household welfare analysis∗ Johannes Haushofer,y Edward Miguel,z Paul Niehaus,x and Michael Walker{ July 6, 2017 Abstract This document outlines outcomes and regression specifications for estimating the effects of unconditional cash transfers on the welfare of recipient households as part of the General Equilibrium Effects (GE) project. This project is a randomized evaluation of an unconditional cash transfer program by the NGO GiveDirectly (GD) in Kenya. This is a two-level randomized controlled trial where treatment status is randomized at the village level and treatment intensity is randomized at the sublocation level, the administrative unit above a village. This document is part of a series of five pre-analysis plans filed to the AEA trial registry as part of the GE project, and focuses on estimating treatment effects for households eligible to receive transfers from GD using data collected via household surveys. We specify regression equations, primary outcomes, and a catalog of outcomes that we will study, as well as corrections for multiple testing and data checks. We also discuss how the analyses specified here fit in with the other pre-analysis plans filed as part of this project. Appendix A: Endline household survey instrument Appendix B: Baseline household survey instrument Appendix C: Household data management note ∗AEA Trial Registry: AEARCTR-0000505, https://www.socialscienceregistry.org/trials/505 This updates a previously-filed version (also from July 6, 2017) to clarify the multiple inference corrections that we will conduct. The rest of the document is unchanged, and the analysis to date section remains accurate. We thank Justin Abraham, Christina Brown, Genevieve Deneoux, and Francis Wong for excellent research assistance, GiveDirectly for collaboration, the Busara Center for Behavioral Economics for survey development and piloting, and IPA-Kenya for data collection. This work has been funded by the Private Enterprise Development in Low-Income Countries (PEDL) initiative, the International Growth Centre, the Weiss Family Foundation, and an anonymous donor. Niehaus is a co-founder and president of GiveDirectly. yPrinceton University, NBER, and Busara Center for Behavioral Economics, [email protected] zUC Berkeley, [email protected] xUC San Diego, [email protected] {UC Berkeley, [email protected] 1 1 Introduction 1.1 Summary This document outlines the analysis plan for endline household survey data collected as part of the General Equilibrium Effects (GE) project, a randomized controlled trial of an unconditional cash transfer program by the NGO GiveDirectly (GD). GD makes large unconditional cash transfers to poor households in Kenya. The magnitude of the transfers is large, around USD 1,000 (nominal) per household, about 75% of annual expenditure for recipient households. At the time of this study, GD targeted households living in homes with grass-thatched roofs, a basic means-test for poverty; we find 33% of households eligible in our study area. (GD currently uses a variety of targeting criteria that distributes transfers to a similar share of households). The intervention involves over USD 11 million in transfers and 653 villages in one Kenyan county. Treatment assignment is randomized at the village level, and within treatment villages, all households meeting GD's eligibilty requirement receive the unconditional cash transfer.1 A second level of randomization provides variation in treatment intensity: sublocations, an administrative unit directly above the village level comprising of an average of ten villages, were randomly assigned to high or low saturation status. In high saturation sublocations, two-thirds of villages were assigned to treatment, while in low saturation sublocations, only one-third of villages were assigned to treatment.2 This analysis plan focuses on estimating direct treatment effects for eligible households (trans- fer recipients). Our experimental design allows us to estimate between-village spillover effects on eligible households, which could contaminate our estimates of the direct treatment effects. This design builds on Haushofer and Shapiro (2016), which allowed for within-village spillovers but as- sumed no spillovers across villages. This research question also relates to a broad literature on cash transfers, which generally finds positive effects for recipient households (Arnold, Conway, and Greenslade 2011 provide a review of the literature). As GD expects to commit around USD 50 million of cash transfers in 2017, estimating direct effects for transfer recipients is highly relevant. This document is part of a series of five pre-analysis plans for the GE project. We already filed two pre-analysis plans: one on midline market price and enterprise phone survey data (Haushofer et al. 2016, filed May 19, 2016.), and one on local public finance (Walker 2017, filed February 12, 2017). We plan to file two additional pre-analysis plans in addition to this plan. One will study the targeting of cash transfers, and seeks to address whether targeting the most deprived households leads to the greatest average treatement effects (this will be referred to as the \targeting" PAP). The other, the general equilibrium (GE) PAP, focuses on prices, output and productivity, and looks into spillovers in more detail. The focus of this document is on direct treatment effects for recipient households, as these are highly relevant to policymakers, and the magnitude of treatment effects will provide context for understanding the results pre-specified in the other analysis plans. 1. This follows GD's typical operating procedure for lump sum transfers. 2. More details can be found in section 1.3 on the experimental design. 2 This pre-analysis plan is not meant to be exhaustive, nor to preclude additional analyses, and we anticipate carrying out additional analyses beyond those described here. The remainder of this document describes the intervention and experimental design (Sections 1.2 and 1.3), the data collected (Section 2), the empirical strategy (Section 3), the primary outcomes (Section 4), and the full catalog of pre-specified outcomes (Section 5). Section 2.1 discusses data examined to date, and plans for examining data once this pre-analysis plan has been filed but before the rest of these pre-analysis plans are filed. The appendix contains household survey instruments and the associated data management note. 1.2 Intervention3 GD provides unconditional cash transfers to poor households in rural Kenya, targeting (for villages in our study) households living in homes with thatched roofs, a basic means-test for poverty. In treatment villages, GD enrolls all households in treatment villages meeting its thatched-roof eligibility criteria (\eligible" households); approximately one-third of all households are eligible. No households in control villages receive transfers. Eligible households enrolled in GD's program receive a series of 3 transfers totaling about USD 1,0004 via the mobile money system M-Pesa.5 This is a one-time program and no additional financial assistance is provided to these households after their final large transfer. GD's enrollment process in treatment villages consists of the following 6 steps: 1. Village meeting (baraza): Before beginning work in a village, GD holds a meeting of all households in the village to inform villagers that GD will be working in their village, explain their program and GD as an organization. To prevent gaming, the eligibility criteria are not disclosed. 2. Census: GD staff conduct a household census of the village, collecting information on house- hold names, contact information and housing materials. The information on housing materials are used to determine program eligibilty. 3. Registration: Households identified as eligible based on the household census are visited by the registration team. GD staff confirm the eligibility of the household, inform the household of their eligibility for the program and register the household for the program. This is the point at which households learn they will be receiving transfers, as well as the amount of the transfers, the transfer schedule, and the fact that the transfer is unconditional.6 Households 3. Some of the text in this section and section 1.3 is reproduced from Walker (2017). 4. The total transfer amount is 87,000 Kenyan Shillings (KES). The average exchange rate from 9/1/14 to 4/30/16 was 97 KES/USD. 5. For more information on M-Pesa, see Mbiti and Weil (2015) and Jack and Suri (2011). 6. To emphasize the unconditional nature of the transfer, households are provided a brochure with many potential uses of the transfer. 3 are instructed to register for M-Pesa, a prerequisite for receiving the transfer. Households that do not have a mobile phone are given the option to purchase one from GD staff, the cost of which is deducted from the transfer amount. 4. Backcheck: All registered households are backchecked to confirm eligibility in advance of the transfers going out. This is an additional step to prevent gaming by households and field staff, as the census, registration and backcheck teams consist of separate staff members. 5. Transfers: The cash is transferred in a series of three payments via M-Pesa according to the following schedule: (i) the token transfer of KES 7,000 (about USD 70) ensures the system is working properly; (ii) two months afterwards, the first lump sum transfer of KES 40,000 is distributed; (iii) six months after this, the second and final lump sum transfer of KES 40,000 is sent. If households elected to receive a mobile phone from GD, the cost of this is taken out of the second lump sum transfer. Transfers are typically sent at one time per month to all households scheduled to receive transfers. 6. Follow-up: After transfers go out, GD staff follow up via phone with transfer recipients to ensure no problems have arisen. In addition, there is a GD help line that recipients can contact. If GD staff learn that household conflicts have arisen as a result of the transfers, transfers were sometimes delayed while these problems were worked out.